Deterministic Sampling for Uncertainty Quantification in Complex Algorithm-Based Measurements

The paper deals with the problem of estimating measured values in indirect measurements based on complex processing algorithm. To this aim, deterministic sampling of the random variables (modeling the input quantities) is suggested to efficiently estimate expectation and standard uncertainty of algorithm outputs. In particular, the authors propose an enhanced version of the traditional unscented transform to propagate a defined set of statistical moments of the input quantities through the algorithms (even in the presence of non-analytical formulation). This way, it is possible to assure estimates of output expectation and standard uncertainty as good as those achieved by means of the very large ensemble of random variates typically exploited in brute force Monte Carlo method.